Deforestation poses a significant threat to global ecological stability, particularly in the Latin American and Caribbean (LAC) region. This study aims to develop an advanced instance segmentation model using deep learning to monitor deforestation with satellite imagery. The model integrates spatial and temporal analysis to accurately identify deforested areas, addressing challenges such as data quality, class imbalance, and varying image exposures with advanced preprocessing, a robust training pipeline, and U-Net and Feature Pyramid Network architecture. A visualization dashboard tracks deforestation over time, enabling model performance evaluation across multiple LAC regions. Validated against known deforestation events, the model effectively detects and monitors forest loss, providing a valuable tool for policymakers and environmental managers.